If you read my list of startups using artificial intelligence to drug discovery, you may have wondered: how much traction do these companies actually have? And perhaps, if you work for a pharmaceutical or biotechnology company, a related question: are any of my competitors working with them?
To help answer such questions, this post summarizes how pharmaceutical companies apply artificial intelligence in drug discovery, including through partnerships with AI startups. As with my startup list, I aim to keep this regularly updated. So if I'm missing anything (which I certainly am, because details of partnerships are often kept secret), or you have news to share, please email me. Or post in the comments.
Here's the list:
- Boehringer Ingelheim
- Bristol-Myers Squibb (BMS)
- Eli Lilly
- Merck Group
- Mitsubishi Tanabe Pharma
- Novo Nordisk
- Ono Pharmaceuticals
- SK Biopharmaceuticals
- Sumitomo Dainippon Pharma
- Wave Life Sciences
- Zambon Pharma
AbbVie has been quiet about its use of artificial intelligence in drug discovery. But it does have a confidential project listed with Atomwise. Also, in September 2016, AbbVie partner AiCure announced how its AI-based patient monitoring platform improved adherence in an AbbVie phase 2 schizophrenia trial.
In May 2018, MIT announced that Amgen was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium. Also that month, a report noted that Amgen is working with medical research machine learning startup Owkin.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Amgen as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
Astellas appears to be focusing AI drug discovery on repurposing existing compounds. It publicized a "Drug Repurposing & Application Management" in February 2015. In December of that year, Astellas and Biovista announced a partnership around drug repurposing. And in January 2016 Astellas and NuMedii announced a similar repurposing collaboration.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Astellas as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
In August 2017, AstraZeneca and Berg Health announced a partnership to discover therapeutic targets for neurological diseases such as Parkinson's.
In February 2018, AstraZeneca announced a partnership with Alibaba to apply technology including artificial intelligence to patient diagnosis and treatment. (They shared few details.)
In April 2019, AstraZeneca reported the start of a long-term collaboration with BenevolentAI. The partnership will focus on new drugs for chronic kidney disease and idiopathic pulmonary fibrosis.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced AstraZeneca as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
Also in June 2019, I received a tip (see comments) that AstraZeneca, via subsidiary MedImmune, had established a multi-year collaboration with ProteinQure. ProteinQure is a fellow Canadian AI drug discovery company that uses quantum computing, molecular simulations, and machine learning to design drugs.
In September 2019, computational drug discovery company Schrodinger announced a partnership with AstraZeneca to deploy a platform that uses physics-based modeling and machine learning to predict the potency of molecules binding to proteins.
In January 2017, BASF announced a partnership with Nuritas, which combines artificial intelligence and genomics to discover natural peptides with health benefits. In November 2018, the pair announced the first peptide to be commercialized under the partnership: a natural ingredient from brown rice that reduces inflammation.
In May 2018, MIT announced that BASF was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium. (Know anything else about what BASF is up to with AI? Please let me know in the comments.)
In May 2018, MIT announced that Bayer was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium.
In August 2018, Bayer included AI drug-design company Cyclica in its Grants4Apps program. In November 2018, Cyclica announced that Bayer was using its technology for off-target effect investigation, pharmacokinetic property prediction, and multi-targeted drug design.
In January 2019, the Alliance for Artificial Intelligence in Healthcare announced Bayer as a founding member.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Bayer as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
In late July 2019, Bayer and Sensyne Health announced a collaboration to develop treatments for cardiovascular disease. Sensyne Health has a unique partnership with the NHS to leverage its electronic patient record data while protecting patient privacy.
One of the earliest AI drug discovery partnerships I could find predates the recent hype about machine learning. Between the Canadian arm of Boehringer Ingelheim and Numerate, in December 2011, it doesn't even refer to AI. A Numerate press release states that it is "leveraging the power of cloud computing and novel computational methods to transform the drug design process." The partnership is focused on generating small molecule drug leads for an unnamed infectious disease target.
In May 2018, news outlets reported that Boehringer had partnered with Bactevo to use its "Totally Integrated Medicines Engine" platform to identify novel small molecule lead compounds.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Boehringer as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
Boehringer is also using AI to analyze speech for signs of neurological disease, to facilitate earlier diagnosis. Their initial focus is schizophrenia and Alzheimer’s disease. A July 2019 report says the company is already using the technology in clinical trials. (It sounds like they're building this in-house. But there are startups out there that focus just on doing this, like Winterlight Labs.)
In February 2018, Bristol-Myers Squibb (BMS) announced entering into a partnership with Sirenas to apply the biotech company's technology to challenging therapeutic targets. Sirenas uses data mining and machine learning to find therapeutic applications of chemicals from global microbiome samples.
In March 2019, BMS announced another partnership, with Concerto HealthAI. Concerto specializes in using AI to analyze real-world oncology data in order to generate insights and real-world evidence. Its partnership with BMS covers a range of data sources, cancers, and activities, including clinical trials, protocol design, and precision oncology treatments.
In January 2019, AI drug discovery startup Exscientia announced that Celgene invested in its Series B. A report in March 2019 says Celgene will have a three-year deal with Exscientia to use its AI in discovering small molecules for three targets.
This said, Bristol-Myers Squibb announced in January 2019 that it was acquiring Celgene. So future AI initiatives may shift under the BMS umbrella.
In May 2018, MIT announced that Lilly was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium.
In July 2018, robotic cloud laboratory provider Transcriptic announced that Lilly would use its AI-powered technology to enable on-demand drug discovery operations.
In June 2019, Atomwise announced a collaboration on up to 10 drugs with Lilly. In the collaboration, it appears that Lilly will use Atomwise's technology to screen molecules that Lilly synthesizes for their therapeutic potential.
Evotec is difficult to describe. It refers to itself as a "drug discovery alliance and development partnership company." Its partners include many more well-known pharmaceutical companies, such as Bayer, Sanofi, Genentech, Janssen, and UCB. It also has a tight partnership with Exscientia.
Evotec announced an initial collaboration with Exscientia in April 2016 and an investment in September 2017. The partnership focuses on creating bispecific small molecule immuno-oncology therapies. These are treatments that can hit two different cancer targets simultaneously. In January 2019, Evotec also announced participating in Exscientia's Series B.
Evotec also partners with academic institutions. It announced a partnership with the University of Oxford in June 2019, for example, to help data-driven drug discovery projects commercialize. For this it will work with clinical artificial intelligence company Sensyne Health.
Unless someone tells me otherwise, Gilead's first publicly announced use of AI in drug discovery was in April 2019. This month, Gilead announced a strategic collaboration with stealthy startup Insitro. The collaboration will focus on nonalcoholic steatohepatitis (NASH). Gilead will use Insitro's platform to create disease models for NASH and find targets that affect the disease's progression and regression.
In June 2017, Genentech and precision medicine startup GNS Healthcare announced a partnership to find and validate potential cancer drug targets by analyzing data from sources such as electronic medical records and next generation sequencing.
In September 2019, Genentech and parent Roche disclosed a predictive analytics project with a paper in Nature on using deep learning to predict which patients with diabetic retinopathy will progress the fastest.
GSK has been one of the most active pharmaceutical companies in applying artificial intelligence to drug discovery. It even created an in-house artificial intelligence unit. (Initially called "Medicines Discovered Using Artificial Intelligence.” Now called “In silico Drug Discovery Unit.”) As of July 2019, GSK's AI team reportedly numbered about 50.
GSK has partnered with startups including Exscientia and Insilico Medicine. The partnership with Excscientia, announced in July 2017, is to discover novel and selective small molecules for up to 10 disease-related targets across undisclosed therapeutic areas. The partnership with Insilico, announced in August 2017, is to identify novel biological targets and pathways.
GSK is also part of the Accelerating Therapeutics for Opportunities in Medicine (ATOM) Consortium, which aims to leverage artificial intelligence to go from drug target to patient-ready therapy in less than a year. (An ambitious goal.) GSK gave ATOM chemical and in vitro biological data for more than 2 million compounds it has screened. (For more information on how GSK became such a leader, check out my article “6 Steps to AI Leadership in Pharma: An Interview with John Baldoni of GSK.”)
In May 2018, GSK announced a partnership to use AI for the design of novel small-molecule drugs with Cloud Pharmaceuticals.
GSK has also worked with Google on applying AI to drug discovery. In July 2018, a paper in PLOS One described how researchers from the two firms developed a machine learning algorithm to identify protein crystals.
In January 2019, the Alliance for Artificial Intelligence in Healthcare announced GSK as a founding member.
In early April 2019, Exscientia announced that its partnership with GSK had produced its first tangible result: a "highly potent" lead molecule targeting a novel pathway for chronic obstructive pulmonary disease.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced GSK as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
GSK also has a collaboration with researchers at the Universities of Strathclyde and Nottingham, announced in July 2019, that focuses on applying AI to synthetic chemistry.
One of the more unique AI drug development partnerships I've seen is that between Janssen and BenevolentAI. In November 2016, they announced that BenevolentAI would license the right to develop, manufacture, and commercialize clinical stage drug candidates from Janssen after using artificial intelligence to identify untapped potential in Janssen's portfolio. This deal may already be bearing fruit, as BenevolentAI recently launched a phase 2b trial for a drug from the partnership to treat sleepiness in people with Parkinson's disease.
In January 2018, Johnson & Johnson Innovation announced a partnership between Janssen and WinterLight Labs to try predicting dementia and neurodegenerative diseases from voice samples obtained through Janssen clinical trials.
In January 2019, the Alliance for Artificial Intelligence in Healthcare announced Janssen as a founding member.
In April 2019, AI-driven drug design startup Iktos announced a collaboration with Janssen. The collaboration will use Iktos' virtual drug design technology on small molecule drug discovery projects.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Janssen as one of its members. (In fact, Janssen issued a press release that was one of the few sources that mentioned all the initial members.) MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
Apart from collaborations to develop new drugs, Janssen is also applying AI to clinical trials. In July 2019, Celsius Therapeutics announced a partnership with Janssen to use its single-cell genomics and machine learning platform to find predictive biomarkers of response in Janssen’s VEGA study of golimumab (Simponi) and guselkumab (Tremfya) in patients with ulcerative colitis.
Like Boehringer Ingelheim, Merck (it's not clear which one) struck an early partnership with Numerate, which they announced in March 2012. The collaboration focuses on generating novel small molecule drug leads for an unnamed cardiovascular disease target.
In December 2018, Merck announced a partnership with Cyclica to use its AI-augmented proteome screening platform to elucidate mechanisms of action, evaluate safety profiles, and explore additional applications for investigational small molecules.
A few months later, in March 2019, Merck announced a partnership with Iktos to use its generative AI to design novel molecules with desired properties for a specified disease.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Merck as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
Mitsubishi Tanabe Pharma partnered with Hitachi to optimize clinical trial planning with AI. Announced in March 2018, the partnership will use Hitachi’s Lumada platform. It will extract information from scientific papers and ClinicalTrials.gov. A pilot of the technology found time savings of 70%.
While not a traditional pharmaceutical company, Nestlé has a health science division. It aims to advance nutrition as therapy. In February 2018, Nestlé announced a partnership with Nuritas to enlist AI in this mission. This partnership will use AI to discover therapeutic peptides in foods.
On the commercial side, Novartis has been quite innovative in using digital media. Its heavily digital campaign for Gilenya, for example, won multiple awards. But until recently, it made relatively few big moves with artificial intelligence. This began to change with incoming CEO Vas Narasimhan. He announced in a September 2017 interview that he planned to partner with or acquire AI and data analytics companies.
To this end, in January 2018, an article revealed that Novartis partnered with McKinsey’s QuantumBlack to analyze clinical trial operations with machine learning. They claim the work has reduced patient enrolment times by 10-15%.
A March 2018 report also describes a partnership with IBM Watson to improve clinical trial recruitment, and the use of a “digital cortex” to predict medication efficacy.
In May 2018, MIT announced that Novartis was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium. This same month, Intel announced working with Novartis to apply AI to high content screening.
In June 2018, Business Insider published an interview with the outspoken Jay Bradner, president of the Novartis Institutes for BioMedical Research (NIBR), about the company's progress with artificial intelligence. Bradner stated that 4% of the 6,000 scientists working at NIBR are data scientists. "We like to think of ourselves as the lead turtle in the race of the turtles," he said, referring to pharma's conservative adoption of emerging technology.
A July 2018 report in InformationWeek highlighted completion of the first phase of Novartis' digital transformation in drug discovery. The company rebuilt its technology infrastructure and connected its data through a single hub. It's now developing a predictive analytics platform that uses machine learning algorithms for clinical trial operations.
In addition to building in-house AI capabilities, Novartis is partnering with academia. In January 2019, the company announced a partnership with the University of Oxford’s Big Data Institute to predict how patients respond to drugs. The work will combine different types of data, such as clinical, imaging, and genomics data.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Novartis as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
In September 2019, Novartis announced a significant move to entrench data science and AI within its business: partnering with Microsoft and creating an AI Innovation Lab. The partnership will initially focus on personalized therapies for macular degeneration, cell and gene therapy, and drug design. A few days later, BenevolentAI announced a partnership with Novartis to use its AI platform for personalizing oncology treatments.
Novo Nordisk included artificial intelligence as a focus of its September 2018 reorganization. The company laid off 400 R&D employees. But it increased investment in artificial intelligence for lead molecule selection and development.
A few months later, in December 2018, Novo announced a deal with UK biotech e-Therapeutics to use its AI-based drug discovery technology to find new treatments for type 2 diabetes. The partnership will include a search for novel intervention strategies, biological pathways, and compounds. It was extended in August 2019.
One of the largest pharmaceutical companies in Japan, Ono Pharmaceuticals in March 2019 announced a collaboration with twoXAR to discover and develop treatments for an unspecified neurological disease.
Of all the companies on this list, only one has publicly promoted a drug discovery partnership using IBM Watson: Pfizer. In December 2016, Pfizer and IBM announced a partnership to accelerate drug discovery in immuno-oncology. There has been little announced since (at least, that I can find), but a string of negative reports about IBM Watson's capabilities (here, here, here, here, here, here, and I could go on), including in healthcare, call into question how fruitful the partnership might have been.
In April 2018, Pfizer was linked to an announcement of funding for UK artificial intelligence initiatives, but the nature of its involvement is unclear.
Pfizer was busy in May 2018, suggesting that it began ramping up AI activity. That month, MIT announced that Pfizer was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium. Pfizer also announced a partnership with XtalPi to combine quantum mechanics and machine learning to predict the properties of drugs. And The Wall Street Journal reported that Pfizer built an internal analytics platform that leverages machine learning to identify patients with rare diseases.
In September 2018, Atomwise announced that Pfizer agreed to evaluate its platform. Under the agreement, Atomwise will generate compounds for up to three target proteins.
In January 2019, CytoReason announced a partnership with Pfizer. Details of the partnership were limited. However, CytoReason's technology uses machine learning to understand how cells respond to diseases and treatments. In previous work, the startup has discovered new cellular players in melanoma, new mechanisms of action in atopic dermatitis, and novel pretreatment biomarkers in inflammatory bowel disease anti-TNFα therapy. CytoReason will standardize and organize Pfizer's data and integrate it into a Pfizer-specific immune system model.
In April 2019, Concerto HealthAI announced a partnership with Pfizer to use AI and real world data in oncology. The partnership aims to find actionable insights for Pfizer's investigational and commercialized therapies for solid tumors and hematologic malignancies.
Roche has made relatively few announcements about its use of AI for drug discovery.
In February 2018, however, Roche acquired Flatiron Health, an oncology-focused electronic health records company. Flatiron's massive amount of oncology data provides Roche with a tremendous asset for machine learning. Also, in May 2018, a report noted that Roche is working with medical research machine learning startup Owkin.
Roche is also one of the few pharmaceutical companies I've seen disclose work with IBM Watson Health. Roche worked with IBM on a study published in January 2019 using real-world data to predict chronic kidney disease in patients with diabetes.
Roche and subsidiary Genentech disclosed another predictive analytics project in September 2019 with a paper in Nature on using deep learning to predict which patients with diabetic retinopathy will progress the fastest.
In April 2016, Sanofi's Genzyme unit and Recursion Pharmaceuticals announced a partnership to use Recursion's drug repurposing platform to screen Sanofi molecules for genetic disease targets.
Sanofi is also another prominent Exscientia partner. Their partnership, announced in May 2017, focuses on finding bispecific small molecule drugs for metabolic diseases such as diabetes and their comorbidities. In August 2019, Exscientia announced that Sanofi exercised its option for a bispecific small molecule targeting inflammation and the progression of fibrosis.
Sanofi and Berg Health also announced a partnership in October 2017 to assess potential biomarkers for seasonal flu vaccine performance.
Sanofi has also experimented with using AI for medical communications. An October 2018 article reported that the company partnered with Researchably, which uses AI to automate medical literature review. Sanofi reportedly found that the software cut the time of reviewing a paper from 13 minutes to less than a second.
In June 2019, Sanofi announced what seems like an extensive partnership with Google, including to leverage AI. Through a new virtual "Innovation Lab," Sanofi and Google will analyze real world data to understand what treatments work for patients, and analyze manufacturing and commercial data to forecast sales and inform marketing and supply chain activities.
In February 2017, Japan's Santen, which focuses on ophthalmic products, and TwoXAR announced a partnership to find new drug candidates for glaucoma.
Another partner of the very active Numerate, Servier and the startup announced in June 2017 a collaboration to design small molecule modulators of ryanodine receptor 2 (RyR2), a target thought to be important in cardiovascular disease that has eluded drug-ability. The collaboration could lead to new treatments for heart failure and arrhythmias.
In January 2019, Servier announced successful results from a collaboration with AI drug design startup Iktos. For the collaboration, Servier used Iktos' technology to analyze 800 molecules designed to meet 11 criteria. None of the molecules met all the criteria. Iktos' technology, however, learned from them. It then generated 150 molecules optimized for the criteria. Servier synthesized and tested them, and found that on average, the molecules met 9 of 11 criteria. One met all 11.
In June 2019, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project announced Servier as one of its members. MELLODDY will train machine learning models on datasets from multiple partners while ensuring the privacy of each partner using federated learning.
In April 2019, SK Biopharmaceuticals, a Korean company that focuses on disorders of the central nervous system and cancer, announced an agreement with twoXAR to develop new treatments for non-small cell lung cancer. Under the agreement, twoXAR will use its AI discovery technology to identify candidates. SK Biopharmaceuticals will then use its internal AI technology to optimize a lead candidate.
In May 2018, MIT announced that Sunovion was a member of its Machine Learning for Pharmaceutical Discovery and Synthesis Consortium.
(Know any more about Sunovion's AI activities? Let me know in the comments.)
One of Exscientia's early partners, Sumitomo Dainippon and the startup announced in September 2015 initial results of a collaboration to identify new treatments for psychiatric diseases. The first compound identified was a bispecific, dual-agonist molecule that selectively activates two GPCR receptors from two distinct families.
Another Numerate partner, Takeda and the startup announced in June 2017 that they would collaborate on identifying candidates for oncology, gastroenterology, and central nervous system disorders.
An August 2018 article reported on Takeda's use of AI to predict the course of depression. Their goal is to determine which patients will improve by switching treatments. Or by starting with different first-line therapies. Initial research showed that recurrent neural networks are accurate at such predictions.
In January 2019, Takeda partner Recursion announced an expanded partnership with the pharmaceutical company to evaluate and identify novel preclinical candidates for rare diseases. The announcement reported that Recursion had evaluated compounds in over 60 indications during the prior 18 months. This reportedly resulted in new therapeutic candidates for more than 6 diseases. Takeda exercised its option for drug candidates in two rare diseases from these early results.
Wave is a biotechnology company focused on genetic conditions with unmet need. In May 2018, Wave and Deep Genomics announced a collaboration to use machine learning to find new treatments for genetic neuromuscular disorders.
Zambon is an investor in precision medicine startup GNS Healthcare. (If you know of any other work Zambon is doing with AI, please let me know in the comments.)
Anything I’ve missed? Let me know in the comments. Add anything I left out or got wrong for the companies above. Or tell me about companies I haven't listed yet.
Image credit: Logos provided by Clearbit